Liquidity Risk And Maturity Management Over The Credit Cycle

Transcription

1 Liquidity Risk And Maturity Management Over The Credit Cycle Atif Mian Princeton University and NBER João A.C. Santos Federal Reserve Bank of New York and Nova School of Business and Economics July 2012 ABSTRACT We use the Shared National Credit data on syndicate loans to investigate U.S. firms refinancing behavior over the last two decades. As credit conditions tighten, refinancing likelihood goes down and draw-down on loan commitments increases sharply. We show that important drivers of the time-series variation in refinancing propensity include maturity management and endogenous demand from more credit-worthy firms. Firms refinance early at a significantly higher rate when credit conditions are good in order to keep the effective maturity of their loans long. Moreover, refinancing propensity is most sensitive to credit market fluctuations for the more credit-worthy firms consistent with the idea that credit-worthy firms choose to refinance at a lower rate when cost of capital rises. * We thank Michael Roberts, Dwight Jaffee, Amir Sufi, Nancy Wallace, Martin Cherkes and seminar participants at U.C. Berkeley (Haas), Federal Reserve Board, NBER Summer Institute and WFA meetings for very helpful comments and suggestions. We thank Vitaly Bord for exceptional research assistance throughout this project. Mian gratefully acknowledges the Coleman Fung Center for funding. The results or views expressed in this study are those of the authors and do not necessarily reflect those of the Board of Governors of the Federal Reserve System or the Federal Reserve Bank of New York or the Coleman Fung Center. Mian: (609) , [email protected]; Santos: (212) , [email protected]. 1

2 A fundamental feature of corporations is that they have long-lived assets while external funding is of limited maturity. Thus firms continually have to go back to banks and renegotiate, or refinance, the maturity of their outstanding loans and credit commitments. The inability of the financial system to guarantee funds for the entire duration of a corporation s life keeps firms susceptible to a sense of fragility: What if banks refuse to rollover their loans or demand a high spread the next time they go for extension? Such concerns expose firms to liquidity risk and costs associated with the necessity to refinance and rollover existing debt at a time when credit is tight 1. A survey of CFOs by Bodnar et al (2011) shows that CFOs worry about managing their maturity structure in order to avoid refinancing in bad times. Figure 1 shows that the cost and ability to refinance a loan can fluctuate wildly over the course of a business cycle. Lending conditions can tighten quickly accompanied by a sudden increase in spreads on refinanced loans. How do firms plan for the prospect of economy wide fluctuations in credit and liquidity risk? Forward-looking firms would want to minimize the possibility that they are forced to refinance loans when lending standards are tough or spreads too high. However, the ability to minimize exposure to economy-wide liquidity risk may depend on certain firm characteristics. Consequently the composition of firms caught in a credit crunch is likely to be systematically different an important fact to keep in mind when estimating the economy wide cost of credit market fluctuations. 1 Holmstrom and Tirole (2011) provide an excellent summary. Some of the relevant literature includes Allen and Gale 2007, DeMarzo and Fishman 2007, Farhi and Tirole. 2011, Gorton and Pennacchi 1990, He and Xiong. 2011, and Krishnamurthy and Vissing-Jorgensen

3 This paper investigates how firms respond to liquidity risk by utilizing a previously unexplored data set on syndicate loans - the Shared National Credit (SNC) program run by the Federal Deposit Insurance Corporation, the Federal Reserve Board, and the Office of the Comptroller of the Currency. The data set follows a syndicate loan over time and tracks if (and when) the loan is refinanced to extend its date of maturity. We can thus track the evolution of refinancing behavior over time and in the cross-section of U.S. corporations. Our data cover the period from 1988 to 2010, enabling us to analyze the behavior of refinancing risk over the course of three business and credit cycles. We find that the refinancing propensity of corporate loans is strongly correlated with the credit cycle. The propensity to refinance a loan is more than fifty percent higher in normal times compared to when credit conditions are tight. Conditional on getting refinanced, a loan s maturity gets extended for longer duration when credit conditions are good. Firms access to unused lines of credit is strongly related to the credit cycle as well. The percentage of drawndown loan commitment increases by seventeen percent when credit conditions are tight. What are the main drivers of the time-series variation in refinancing propensity and its correlation with the credit cycle? On the credit-supply side, one would expect that when banks cut back on liquidity in tough times, it is the weaker firms i.e. firms deemed to be less credit worthy that are most likely to see a reduction in their refinancing propensity. However, we find that completely the opposite is true. It is the refinancing propensity of credit worthy firms that declines the most when credit conditions tighten. For example, refinancing propensity of investment grade firms is almost four times as sensitive to credit market conditions as non-investment grade firms. Using lead bank- 3

4 year fixed effects, among other controls, we show that this result is not driven by any lead-bank specific credit supply shocks. Thus the significantly higher sensitivity of refinancing propensity to credit market conditions for credit worthy firms is likely to be driven by demand-side dynamics. We show that an important demand-side channel for the sensitivity of refinancing to credit market conditions is dynamic maturity management: firms refinance early and extend the maturity of loans during normal times well before these loans become due. Firms that are successful at doing so have a greater capacity to ride out liquidity shocks by not having to refinance when credit costs are high. The maturity management effect is strong enough to affect the maturity of outstanding loans over the credit cycle. In particular, firms actively manage their maturity structure through early refinancing of outstanding loans. The effective maturity of loans increases by about sixteen percent when refinancing is high in times of cheaper credit. Sixty five percent of loans that get refinanced do so with over a year still left in existing maturity - forty percent do so with over two years left in existing maturity. The pattern of early refinancing is not constant through time but displays a distinct cyclical pattern. The relative propensity (hazard ratio) to refinance early versus at-maturity increases by over fifty percent when credit conditions are good. Moreover, loans that are refinanced early are systematically more credit-worthy. We show that early refinancing propensity is strongly related to firm fundamentals including sales growth, credit rating, access to public equity, and excess debt capacity 2. Consequently, firms that do not refinance their loans 2 Excess debt capacity is measured as percentage of total loan commitments that remain unused by a firm. 4

5 until the last year of maturity and hence are most exposed to rollover risk are systematically weaker firms. A possible alternative explanation for our finding is that there is no active maturity management done by credit worthy firms, but instead their natural demand for credit is more cyclical. We find this unlikely a priori because less credit worthy firms are more likely to be high beta in general and display a more pro-cyclical properties. The fact that we find evidence to the contrary is again suggestive of the power of maturity management. Finally, even when we condition on loans that are in the last year of their maturity, we find that loans from credit-worthy firms are more sensitive to credit market fluctuations. For example, refinancing propensity of non-investment grade loans in their last year of maturity is statistically insensitive to credit market conditions, while that of investment grade loans is negatively correlated with credit market tightness. Our results highlight the role that maturity management plays in reducing exposure to liquidity risk. Since credit-worthy firms are the ones successful at doing so, our results also emphasize the endogenous nature of firms caught in a liquidity squeeze. Firms that are unable to maintain longer maturity debt during normal times and hence more likely to be forced to rollover debt in tight credit conditions are systematically worse quality firms. Our finding that maturity extension through early refinancing is closely linked to credit worthiness is consistent with theoretical work such as Flannery (1986), Diamond (1991), Hart and Moore (1994), and Berglof and von Thadden (1994) that argues that banks would deliberately want to keep maturity structure short in order to gain more leverage and control visà-vis less credit worthy firms. Our finding that credit worthy firms minimize the need to be 5

6 forced to refinance in tough times is consistent with risk management literature such as Froot, Scharfstein and Stein (1993). Such maturity risk management is also useful from a macro perspective since it lowers potential mismatches between liquidity supply and demand in times of trouble. There have been a number of empirical studies on the determinants of overall corporate debt maturity (Barclays and Smith (1995), Stohs and Mauer (1996), Guedes and Opler (1996), Johnson (2003), and Berger et al (2005)). These studies primarily focus on the cross-sectional relationships between a firm s characteristics and its choice of corporate debt maturity. Our paper in contrast focuses on the dynamic refinancing choice and its relationship with business cycle and firm credit worthiness. A recent related paper is Almeida, Acharya and Campello (2011) who show that low beta firms manage their liquidity through bank credit lines while high beta firms prefer cash. If we consider low beta firms as more credit worthy, then our results imply that low beta firms can afford to rely more on credit lines because they are better able to manage the maturity risk via early refinancing of loans in good times. Our paper is closest in spirit to Roberts and Sufi (2009) who use SEC filings to show that renegotiation of syndicate loans happens often and is mostly voluntary driven by improvements in credit quality and credit market conditions. Our focus is more on the inter-play between refinancing choice at the firm level and credit / business cycle dynamics as our data covers three recessions and follows loans until the end of their maturity period. The focus of our paper is on the relationship between maturity extension and credit market conditions. A related paper by Roberts (2010) focuses on renegotiation of other loan 6

7 terms / covenants as well and shows that loans where the initial contract is deliberately restrictive due to high information asymmetries are more likely to renegotiate more frequently. The rest of the paper is organized as follows. Section I describes the data and presents summary statistics. Section II presents aggregate trends in credit conditions, refinancing propensity, and maturity structure. Section III presents the main empirical results while section IV discusses the possible interpretations of our results and section V concludes. I. Data and Summary Statistics A. Data Our main data source for this project is the Shared National Credit (SNC) program run by the Federal Deposit Insurance Corporation, the Federal Reserve Board, and the Office of the Comptroller of the Currency. 3 The SNC program gathers, at the end of each year, confidential information on all credits - new as well as credits originated in previous years - that exceed $20 million and are held by three or more federally supervised institutions. For each credit, the program reports the identity of the borrower, the type of the credit (e.g. term loan, credit line), its purpose (e.g. working capital, mergers and acquisitions), origination amount, origination date, maturity date, rating, and information about the syndicate. The program reports both the outstanding amount on a loan, as well as the total loan commitment that the borrower may withdraw. The SNC data not only reports the total commitment of a syndicate loan, but also breaks down this loan commitment by lead bank and all of the participant banks in the syndicate. We thus know the identity of all participating banks in a syndicate, as well as their relative share in the total loan. 3 The confidential data were processed solely within the Federal Reserve for the analysis presented in this paper. 7

8 Since the SNC program gathers information on each syndicate loan at the end of every year (December 31 st ), we can link loans over time and construct variables that capture changes in loan terms (such as maturity date or loan commitment) as well as changes in the amounts drawndown by borrowers each year. Similarly, we can follow the performance of loans over time in terms of credit ratings. Earlier studies of the syndicate loan market (see e.g. Sufi (2007) and Santos and Winton (2008)) use loan origination data from DealScan. An important advantage that the SNC data set has over Dealscan is that it has complete information on the share of the credit held by each member of the syndicate. Another important advantage of the SNC data set is that it tracks each syndicate loan over time. This gives us the opportunity to track both the performance as well as refinancing behavior of loans over time. We also follow the performance of borrowers that are publicly listed by matching our SNC data with financials data from Compustat and stock price data from CRSP. On the lender side we merge data on bank financials for the lead bank. This data come from the Reports of Condition and Income compiled by the FDIC, the Comptroller of the Currency, and the Federal Reserve System. The data include the bank's capital-to-asset ratio, its size, profitability and losses / charge-offs. Wherever possible we obtain bank data at the holding company level using the Y9C reports. If these reports are not available then we rely on Call Reports which have data at the bank level. Table 1 tabulates the basic description of the SNC data. The data covers 50,469 unique syndicate loans over 1988 to 2010 for a total of 156,041 loan-year observations (column 1). Our unit of analysis in this paper is the loan-year. While the coverage of SNC loans increases over time, on average we have four to eight thousand syndicate loans in a given year. A syndicate 8

9 loan may disappear from the SNC data set over time if the size of the loan is no longer large enough to warrant reporting by the lead bank or if it is not held by at least three federally supervised institutions. While we are cognizant of this potential incompleteness in our panel, we do not believe it biases the core results of our paper in any obvious direction. There are a total of 22,156 distinct corporate firms (borrowers) represented in our data with 3,312 to 5,360 firms in any given year (column 2). Some of our tests focus on firms with multiple loans in the same year, such that the loans have different number of years left till maturity. Column (4) reports the number of such firms every year. In total there are 5,749 firmyears that satisfy this constraint. The number of lead banks varies from 305 to 163 over the sample period with a total of 661 unique banks. Finally, column 6 reports the distribution of industries in our sample, with manufacturing being the most represented industry. B. Summary Statistics The top panel in Table 2 characterizes our sample of syndicate loans. The average loan commitment is 188 million dollars, with the 10 th and 90 th percentile being 15 million and 409 million respectively. Thus our data covers large corporate loans. The average outstanding loan is about half the amount of average commitment as the average draw down percentage is 57 percent. 84 percent of loans have an investment grade rating. On average, lead banks lend 23 percent of the syndicate loan, 20 percent of lead banks are foreign, and 32 percent of a syndicate loan is funded by shadow banks - defined as non-commercial financial institutions. A key variable of interest in our paper is whether a loan gets refinanced at a point in time. We construct the average propensity to refinance in the following manner. A syndicate loan i is defined to be refinanced in year t if its date of maturity at the end of year t is greater than the date 9

10 of maturity for the same loan at the end of year t-1. In the event a loan is observed at the end of year t-1 but not observed later on, we assume that the loan was not refinanced 4. Since it is possible for loans to sometimes drop out of our sample for reasons mentioned before, our definition of refinancing underestimates the level of true refinancing. However, we are mostly interested in the time-series variation in refinancing likelihood, and there is no particular reason to think that the cyclical pattern would be biased in any direction due to our variable construction. The unconditional refinancing probability is 21.7%. The upper-left, upper-right and lower-left panels of Figure 2 plot the distribution of maturity structure and changes in maturity structure for syndicate loans. The upper-left panel shows that close to eighty percent of the time, there is no change in the maturity date of a loan. However, conditional on a change in maturity date, it is mostly extended by one year followed by two and three years respectively. The modal maturity of loans at origination is five years, but maturities of up to seven years at origination are fairly common (lower-left panel). Since remaining maturity declines over time after origination, the distribution of maturity left is shifted to the left in the upper-right panel. It is also smoothed out since maturity left is measured as of December 31 st of each year, and loans are originated throughout the year. The lower-right panel shows the draw-down percentage distribution is bi-model. Onethird of loans are fully drawn down, while one-quarter of loans have not been utilized at all. The distribution is fairly uniform within these two extremes. In the analysis that follows, we will make important use of the information that some firms are borrowing up to their maximum 4 If a loan is refinanced by a different lead bank, our data allow us to track that change and we code it as a refinancing. Also, note that the option to refinance a loan may be embedded into a loan contract upfront at the time of origination. However, we do not know if a loan has such embedded refinancing options. 10

11 capacity and thus may be credit constrained. This is a novel feature of our data set that we can observe total commitments as well as how much firms draw down against these commitments. The middle panel of Table 2 presents firm financials for the subset of borrowers that are publicly traded. The average firm has assets worth 3.5 billion dollars, with total sales worth 2.4 billion dollars. The average growth in sales is 15 percent and the average return on assets is 2 percent. The lower panel of Table 2 presents summary statistic for our key macro variables (at annual frequency) measuring credit cycle and business cycle conditions. Business cycle strength is measured by GDP growth while credit cycle strength is measured by senior loan officer survey responses to questions regarding loan spread increase and lending criteria tightening. The survey questionnaire has a scale of -100 to 100 that we re-normalize after dividing by 100. The number of observations is less for the senior loan officer survey as the survey starts in The correlation between loan spreads and loan tightening response is very high at 91.8%. GDP growth and credit conditions tightening are negatively correlated. The correlation is -22.1% at an annual frequency, and increases to -57.5% at quarterly frequency. II. Liquidity And The Credit Cycle: Aggregate Patterns A. Credit Conditions We begin by highlighting aggregate trends in banking sector credit conditions. The top panels in Figure 1 plot the average response to loan officers survey on credit tightening for large and medium C&I loans. The dashed vertical lines represent recessions as dated by NBER. The upperleft panel shows that loan officers consistently report that they have tightened lending 5 GDP growth is from 1989 to 2010 and senior loan officer data from 1990 to is not included in GDP growth summary statistics because we need one year of pre data in running our regressions. 11

12 standards around recessions. However, sometimes (e.g. 1998) credit conditions tighten significantly without corresponding drop in GDP growth. The upper-right panel reports the senior loan officer responses to questions regarding increases in spreads. The plot is very similar to the upper panel with a quarterly correlation of 89.5%. The overall credit tightening at the aggregate level could either be driven by supply-side conditions for example due to losses to bank capital seen earlier or by demand-side conditions such as greater uncertainty about firms future cash flows. The aggregate picture is insufficient to separate supply and demand side forces. For example, the lower-left panel in Figure 1 shows that bank chargeoffs rise during recessions. This could adversely impact bank capital and loan supply. However, the lower-right panel shows that interest rate spreads also rise during recessions. The rise in spreads could derive from a deterioration in banks financial condition (Santos (20011)), or it could reflect tougher cash flow position for firms going forward and hence reduced credit worthiness for borrowers. 6 B. Corporate Liquidity A unique feature of our data is that we directly observe the refinancing likelihood of a loan at a point in time, which is often the key variable of interest in the theoretical literature on corporate liquidity. The construction of this variable is already discussed in section II. Figure 3 plots the evolution of refinancing probability over time. The propensity to refinance a loan shows strong cyclical properties. Refinancing probability is 17.1% on average during recession years and when credit conditions are tight, 6 Santos (2011) shows that banks that experienced larger losses during the subprime crisis demanded higher spreads on the loans they extended to corporations. 12

13 while it reaches 30.7% and 28% respectively at the peak. These refinancing probabilities are not conditioned on the time left in current maturity, and combine both term loans and credit lines. While we will separately condition on years left till maturity later on, the time-series refinancing pattern looks similar for both credit lines and term loans 7. In the appendix (appendix figure 1) we plot the average number of years that maturity is extended by conditional on a loan being refinanced. The size of maturity extension conditional on refinancing also displays some cyclicality. Appendix figure 1 also plots the share of outstanding loans at the end of a given year that is new loans. While our focus in this paper is on refinancing and not first time loans issues, the percentage of loans that are new in a given year is also pro-cyclical. During recessions, only 18.6% of loans are new loans issued in those years for the first time. This proportion reaches 25.1% and 25.7% at peaks. C. Undrawn Loan Commitments While the ability to refinance an outstanding loan is an important measure of liquidity for firms, another relevant metric is the availability of unused lines of credit. These are loan commitments that firms can tap into in case of a sudden liquidity need. An important feature of our data set is that we observe not just the outstanding loan amount, but also the total loan commitment that banks have issued. The top panel in figure 4 plots the average percentage of total loan commitment that is drawn down by a firm. The draw down percentage is pro-cyclical and varies strongly with credit conditions. While the average draw down percentage is 60.8% during recessions, it reaches as 7 The key distinction between credit lines and term loans is their maturity at origination, a variable that we will explicitly account for in our analysis later on. 13

14 low as 53.3% and 50.3% in normal years. The cyclical pattern in draw down percentage is present through all of the three business and credit during our sample period. While some have pointed out to the sharp increase in corporate drawn down rates in the most recent recession (Ivashina and Scharfstein (2010)), our results show that this pattern is representative of previous recessions as well. In particular, the draw down percentage begins to rise before the onset of a recession in all three recessions. The rise is also strongly related to credit market conditions. For example, drawn down percentage rises sharply in 1998 and beyond when credit conditions tighten significantly. The percentage of drawn down is not uniform across all loans. As figure 2 demonstrated, some loans are not drawn upon at all, while others are maxed out. Loans that are fully drawn may be of particular interest as they potentially reflect firms facing financial constraints. The lower panel plots the percentage of loans that are fully drawn out over time 8. As with the average draw down ratio, the percentage of loans that are fully drawn down is closely related to business and credit cycles as well. Interestingly the percentage of loans that are fully drawn out is higher in the 2001 recession than in the recession. D. Effective Maturity As new loans get issued and old loans get refinanced, the overall maturity structure of outstanding syndicate loans constantly changes. The average effective maturity of all outstanding syndicate loans is of interest from a liquidity risk perspective since shorter maturities indicate greater susceptibility to financial fragility. Figure 5 uses information on the date of maturity for each outstanding loan as of December 31 st of each year to plot the average effective maturity over time. 8 Loans with draw down percentage greater than 95% are defined as fully drawn out. 14

15 The upper-panel plot shows that there is a trend as well as a cycle in the evolution of average maturity over time. There is an unmistakable downward trend in average maturity of outstanding syndicate loans over time. While average maturity is close to four years in 1988, it declines to just over two and a half years in This drop of almost a year and a half in effective maturity should be of independent interest. For example, one possibility worth exploring is that increased reliance on short term borrowing (such as Repo transactions) forced banks to favor shorter term lending over time 9. Of more immediate interest for us is the embedded cyclicality in average maturity over time. If we take out the downward trend, there remains a strong cyclical component such that average maturity declines by about six months during recessions. The decline in average maturity during recessions may be driven by two separate factors. First, banks may issue new loans of shorter maturity. We find that the maturity of new loans issued does decline during recessions. Second, borrowers may be more likely to refinance early in non-recession years and thus extend the maturity of their outstanding loans. As we show in more detail in subsequent analysis, this second channel is very much operative as well. One consequence of greater likelihood to refinance early in non-recession years is that fewer percentage of loans reach their maturity limit in non-recession years. The lower-panel in figure 5 shows that the percentage of loans that reach their maturity limit almost doubles in recession years relative to non-recession years. III. Why Is Corporate Liquidity Pro-Cyclical? Section II shows that credit conditions and corporate liquidity are strongly correlated. In 9 Brunnermeier and Oehmke (2011) discuss how a maturity rat race between banks could lead to declining maturity for bank funding. 15

16 particular, as credit conditions tighten and risk spreads widen, corporate liquidity as measured by the propensity to refinance an existing loan and the availability of unused lines of credit goes down. What drives the time-series variation in corporate liquidity as measured by refinancing propensity? We begin by documenting how refinancing propensity depends on borrower attributes in the cross-section. As we will see later, the change in dependence of refinancing propensity on borrower characteristics over time is instrumental in understanding the sources of time-series variation in corporate liquidity. A. Refinancing And Loan Characteristics In The Cross-Section We test how the probability of refinancing depends on borrower characteristics by estimating the following regression specification.,, 1 where Y is an indicator variable for whether the syndicate loan i of borrower j from lead bank b is refinanced during year t. According to equation (1), the probability of refinancing may depend on loan characteristics and borrower characteristics. Since we want to focus our attention to the dependence of refinancing on loan and borrower characteristics, we non-parametrically absorb any time-series fixed effects and any bank-specific shocks by including bank-year fixed effects. While the left hand side variable captures whether a loan is refinanced during year t, all of the right hand side variables are measured as of the end of year t-1. Equation (1) may be estimated using a non-linear procedure such as logit. However, for simplicity and ease of interpreting coefficients, we present our results using the linear probability model. Nonetheless all of our results are robust to using a logit estimation procedure. Column (1) of Table 3 estimates equation (1) without bank-year fixed effects. The result 16

17 indicates that loans with high credit rating, loan commitments that are not drawn to the maximum, credit lines, loans of publicly listed companies, loans with fewer years left till maturity, and loans with shorter maturity at origination are all more likely to get refinanced. Even without any time fixed effects, these loan and borrower level attributes collectively explain ten percent of the variation in the left hand side indicator variable. These results suggest that loan level attributes are quite important for determining the likelihood of refinancing. The magnitude of these coefficients is quite large as well. While the average propensity to refinance a loan is around twenty one percent (Table 2), loans with an investment grade rating are 6.7 percentage points more likely to get refinanced. Loan commitments that are drawn down more are less likely to be refinanced, and this effect is non-linear with the refinancing probability dropping suddenly for loans that are fully drawn down. Credit lines are 2.6 percentage points more likely to get refinanced. Firms with access to public equity are 4.1 percentage points more likely to get refinanced. Finally, as loans get closer to maturity date, they are more likely to get refinanced. Column (2) includes year fixed effects to control for average time variation in refinancing probability and thus only exploits cross-sectional variation in borrower attributes to identify coefficients of interest. It also includes separate fixed effects for number of years left in maturity, and number of years for maturity at origination. The fixed effects thus non-parametrically control for any effect of maturity structure on refinancing likelihood. We also add industry fixed effects to control for variation driven by industry differences. While the R-square increases to 0.17 with the addition of various fixed effects, our coefficients of interest remain largely unaffected. The effect of draw down percentage is now even more non-linear, with refinancing probability dropping discontinuously by 3.5 percentage 17

18 points for loans that are fully drawn out. This result suggests that loans that are completely drawn out are fundamentally different from other loans and likely reflect that such borrowers are financially constrained. We shall explore this classification further in later tests. Column (3) repeats the analysis for column (2), but restricts the analysis to borrowers that are publicly listed. Doing so brings the added advantage that we can now include variables such as change in log sales, return on assets and sales over assets that capture borrower performance. The coefficients on loan level attributes are similar, with the effect of credit rating and financial constraints being even more important than before. Of the new firm performance variables added, only growth in sales comes in significant. A one standard deviation increase in sales growth increases the refinancing probability by 0.64 percentage points. Columns (4) and (5) add bank-year fixed effects to columns (2) and (3) respectively. Bank-year fixed effects absorb any time-varying shocks at the lead bank level that might affect the refinancing probability. For example, if there are certain bank-specific credit supply shocks such as hits to bank capital, these are completely absorbed through the bank-year fixed effects. While the R-square increases by about five to eight percent, the coefficients on borrower and loan characteristics are materially unaffected. Columns (6) and (7) replicate columns (4) and (5), but restrict analysis to loans that have less than one year left till maturity. These are loans that are under greater pressure to get refinanced. We find that borrower attributes matter more strongly for this set of loans. In particular, the effect of credit rating and of commitment being fully drawn out almost doubles. Overall the results in Table 3 show the importance of loan and borrower characteristics in determining refinancing likelihood. Since we include bank-year fixed effects, none of the estimated coefficients are driven by supply-side shocks to individual banks. 18

19 B. Refinancing Early Versus At-Maturity Table 3 also shows not surprisingly that propensity to refinance increases as loans get closer to maturity date. The top-panel in figure 6 analyzes this relationship further by separating the probability to get refinanced by the number of years left till maturity (blue-solid line). The average probability to get refinanced is over fifty percent for loans with less than a year left till maturity, and declines as number of years till maturity increases. However, refinancing probability remains significant even for loans with multiple years left till maturity. We classify refinancing of loans that have more than a year left in maturity as early refinancing, and refinancing of loans with less than a year left in maturity as at-maturity refinancing. The red-dash line plots the share of refinancings that are done for loans with a given number of years left till maturity. Forty five percent of refinancings are done at-maturity and the remaining fifty five percent are early refinancings. While the average rate of refinancing is higher for loans at-maturity, the results from Table 3 suggest that loans that remain non-refinanced until the last year of their maturity are systematically different from loans that get refinanced early. In particular, loans that reach maturity before being considered for refinancing are likely to be loans with lower credit worthiness. Since the worse credit worthiness lowers the propensity to refinance on average, an implication of the selection effect for loans refinanced early versus at-maturity is that the gradient of the solid blue line in the top-panel of figure 6 is biased downwards. We test for this by first estimating a version of the solid blue line of top-panel in a regression framework by estimating the following OLS regression equation and plotting the coefficients on indicator variables for years left till maturity: 19

20 2 where Y is refinancing indicator variable, is a vector of indicator variables (fixed effects) that turn 1 if the number of years left till maturity is between n and n+1. n varies from 0 to 10, with all loans above 10 years left in maturity top-coded at 10. represent fixed effects for the number of years of maturity at origination of a loan (also going from 0 to 10), and are year fixed effects 10. The dotted blue line in lower-panel of Figure 6 plots the coefficients. We next reestimate these coefficients after adding borrower-year ( ) fixed effects in equation (2). Borrower-year fixed effects force comparison to be within-firm in a given year. It thus compares within firm-year differences in refinancing rates for borrowers that have multiple loans such that one loan is maturing in the coming year, and another loan has more than one year left till maturity. Coefficients with borrower-year fixed effects added to equation (2) are thus immune from the borrower selection effects mentioned earlier. Our strategy is similar to the firm-year fixed effects strategy introduced by Khwaja and Mian (2008) to isolate the impact of credit supply shocks. The dashed red line in lower panel plots the coefficients with borrower-year fixed effects. As expected the gradient of the red line is higher than the gradient of OLS coefficients, consistent with the notion that OLS gradient is biased downwards due to unobserved borrower heterogeneity. C. Why Is Corporate Liquidity Pro-Cyclical: The Role Of Maturity Management 10 For exposition clarity, we suppress coefficient estimates of fixed effects from the specification whenever these coefficients are not reported in tables. 20

21 We now turn towards understanding the drivers of time-series variation in corporate liquidity, as defined by the refinancing propensity. Since the cost of credit can rise sharply and unexpectedly when credit conditions tighten, firms would want to plan ahead and keep the maturity of their loans as long as possible. The evidence in figure 6 has shown that a large fraction of loans are refinanced early. We now investigate how the propensity to refinance early varies with the credit cycle. The top panel in figure 7 breaks down the time variation in probability of refinancing by loans that are maturing within the next year, i.e. at-maturity refinancing (left axis) and loans that are maturing after more than one year, i.e. early refinancing (right axis). While both early and atmaturity refinancing likelihoods vary with the credit cycle, early refinancing is more cyclical than at-maturity refinancing. For example, when credit conditions tighten in 1998 and remain tight for a few years (see figure 2), early refinancing likelihood drops more aggressively over the time period. Importantly, while the average probability is much lower for early refinancing versus atmaturity refinancing (16.6% versus 52.2%), the swings in refinancing likelihood for early refinancing are larger in absolute terms. Thus in proportionate terms (or in terms of the odds ratio), early refinancing is a lot more sensitive to business cycle fluctuations than at-maturity refinancing. This can be seen more clearly from the lower panel that explicitly estimates the odds ratio of refinancing early versus at-maturity. This odds ratio is estimated using the following logit model separately for each year: 1 Pr

22 where LM is an indicator variable for long maturity syndicate loans, defined as loans with more than one year left till maturity. are fixed effects for the number of years of maturity of a loan at origination. thus compares the early versus at-maturity refinancing likelihood for loans that had the same maturity at origination but differ in effective maturity because of the time when they were originated. Equation (3) is estimated separately for each year. The blue dashed-dotted line in the lower panel of figure 7 plots the estimated. Given the logit specification, represents the log of the odds ratio for early versus at-maturity refinancing over time 11. The odds ratio is correlated with credit conditions and the magnitude of cyclicality is large as well. As we mentioned earlier, one concern with the odds ratio estimate is that loans refinanced at-maturity are likely to be different in unobservable ways to loans that are refinanced early. While this may bias the level of the odds ratio estimate, it is not clear how such a bias would impact the cyclicality of the odds ratio with respect to the credit cycle. Nonetheless as outlined before, we can adopt the approach that only considers borrowers with loans of multiple maturities in a given year such that one of the loans is maturing in the coming year (i.e. is atmaturity) and another loan is maturing with more than one year left in maturity. Restricting our analysis to this subsample guarantees that early versus at-maturity loans are coming from the same borrower on average. The red dashed line in the lower panel of Figure 7 plots the year-by-year estimate of after restricting the sample to borrowers that satisfy the above multiple maturity criteria. The resulting graph shows that once any inherent bias in early refinancing is removed, odds ratio becomes even more cyclical with respect to credit conditions. 11 Since we are interested in estimating the odds ratio, we switched to a logit model in this section instead of our usual choice of linear probability models. 22

23 Table 4 formally tests for the correlation of early versus at-maturity odds ratio by interaction the LM indicator variable with measures of credit cycle and business cycle strength (MACRO). Equation (3) is thus updated to: 1 Pr Equation (4) adds year fixed effects ( ) since the estimate is done over the entire sample rather than each year separately as in (3) and industry fixed effects ( ). The coefficient of interest is that captures how the odds ratio is correlated with macro conditions. Standard errors are clustered at the lead bank level as always. There are a total of 649 distinct lead banks over the entire regression sample. Column (1) presents the average odds ratio over the entire sample without including macroeconomic variables. In order to convert the estimated coefficient into an odds ratio, we need to take its exponent, i.e., in Table 4. The estimated odds ratio is 0.26 and very tightly estimated. Thus on average, the probability of early refinancing is 26% that of at-maturity refinancing. This result is materially unaffected if we add our previous controls to column (1), i.e. credit rating, drawn down percentage, fully draw down indicator variable, indicator variable for credit line, and an indicator variable for whether the borrower is publicly listed. Column (2) tests directly for the cyclical properties of the odds ratio by interacting the long maturity indicator variable with indicator variables for each recession separately. The results indicate that odds ratio is significantly lower in both the 2001 and recessions, with the drop in the recession being the largest. The odds ratio is actually higher in the 1991 recession, but the magnitude of increase is small and statistically not significant. Columns (3) and (4) test directly for the correlation of the odds ratio with credit conditions as reflected by results from loan officers surveys on credit tightening and credit 23

24 spreads. The negative and statistically significant coefficient on the surveys shows that the relative odds of refinancing early decrease as credit is tightened or as credit spread increases. Using column (3) estimate, a one standard deviation increase in credit tightening (0.23) leads to a 20.7 percent decline in odds ratio of refinancing early. Column (5) replaces credit condition variables with GDP growth that captures overall business cycle conditions. The interaction is positive and statistically significant. A one standard deviation decline in GDP growth (1.83) leads to a 16.5% decline in the odds ratio of refinancing early. The business cycle and credit cycle are naturally highly correlated. Column (6) includes both credit condition and GDP growth interactions in the same regression and finds that the coefficient on credit conditions interaction remains significant and strong in terms of magnitude. This result can also be visually glanced from the lower panel of figure 7: The odds ratio for early refinancing declines sharply when credit conditions tighten (as in 1998) even if GDP growth does not suffer an equivalent decline. Finally, column (7) restricts our sample to borrowers with multiple loans in a given year, such that one loan is maturing in the coming year and another sometime afterwards. Doing so corrects for the concern that loans maturing within the next year are systematically coming from worst quality firms. With the sample restriction, the interaction term with respect to credit conditions becomes even stronger. Thus for the same set of firms, their decision to refinance early versus at maturity is strongly related to credit conditions. As before, this result is not driven by a spurious correlation with business cycle strength as measured by GDP growth. Adding GDP growth interaction to column (7) increases the coefficient from to

25 The results in table 4 and figure 7 help understand the cyclical variation in effective maturity observed in figure 5. When credit conditions are relaxed and cost of credit is low, firms try to refinance early and keep as long a maturity as possible. Such time-varying maturity management reduces the probability that firms will be forced to refinance their loans at a time when credit is costly. Could it be that the cyclical variation in refinancing overall, and early-refinancing in particular is driven by the credit supply side? For example, perhaps firms want to refinance at the same frequency all the time, but the time variation in refinancing rate is entirely driven by reduction in refinancing propensity from the supply side. We address this question next by testing for one particular comparative static prediction of the supply-side effect. If the time-series reduction in refinancing propensity when credit conditions tighten is driven by banks refusal to refinance at the same rate as before, then one would expect the reduction in refinancing rate to be stronger for weaker firms. In particular, if banks have to let some borrowers go due to supply-side constraints, then they should let go of the marginal and less credit-worthy borrowers first. We test this prediction in the next section. D. Why Is Corporate Liquidity Pro-Cyclical: The Role of Demand-Side Forces We test how the refinancing pattern of syndicate loans over time varies by borrower credit worthiness. To do so, we first regress refinancing indicator variable (0/1) on fixed effects for the number of years left till maturity, and fixed effects for the number of years of maturity at origination of a loan. The residual from this regression has thus been stripped of any fixed differences in refinancing propensity driven by the maturity structure of a loan either at origination or at the time of observation. 25

26 Figure 8 plots this residual over time separately for various borrower categories 12. The top panel plots the refinancing likelihood stripped of the effect of maturity structure separately for loans that are drawn down to the maximum (red dotted line), and loans that have only used less than five percent of their total loan commitment (blue solid line). The comparison thus focuses on borrowers that are likely to be financially constrained (the red dotted line) and borrowers with excess debt capacity (blue solid line). The results indicate that borrowers with excess debt capacity are more pro-cyclical than financially constrained borrowers. For example, while financially unconstrained borrowers have refinancing likelihood that is five to ten percentage points higher, refinancing likelihoods become identical for both type of firms in the recession! Similarly, there is a sharp drop in refinancing likelihood for borrowers with financial slack in the wake of the 1998 liquidity crisis, suggesting that more credit worthy firms deliberately cut back on refinancing loans when liquidity costs rise. The middle panel in figure 8 compares loans with an investment grade rating with noninvestment grade loans. As before, investment grade loans are more sensitive to credit cycle costs. In fact most of the overall time-variation in refinancing likelihood is driven by investment grade loans. The bottom panel of figure 8 compares loans from borrowers with access to public equity (blue solid line) with loans from borrowers that are private (red dashed line). Once again borrowers with greater access to external financing display greater cyclical tendency in terms of refinancing probability. 12 The shape of graphs in Figure 8 is similar if we do not strip away the maturity structure effect first. However, conceptually we want to focus on the variation that is not driven by differences in the maturity structure. We therefore show results after stripping away the fixed effect of maturity structure. Appendix figure 2 reports time variation in early refinancing propensity by credit worthiness and shows that as in figure 8, early refinancing by credit worthy borrowers is more sensitive to credit market conditions. 26

27 The results in Figure 8 suggest that the time variation in refinancing likelihood is driven by demand-side forces. In particular, credit worthy firms deliberately choose not to refinance when credit standards tighten and loan officers demand higher spreads. It is hard to imagine scenarios under which the supply-side (i.e. banks) would impose a harsher treatment on more credit-worthy borrowers in weak economic times 13. We test more explicitly for how the sensitivity of refinancing likelihood in the crosssection varies with borrower fundamentals over time by estimating the following regression specification: 5 where is some measure of borrower credit worthiness (fundamentals) such as credit rating, percentage of loan commitment that is drawn out, and whether borrower has access to equity markets. The coefficient captures the sensitivity of refinancing likelihood to these borrower fundamentals.,, denote fixed effects for number of years left in maturity, number of years of maturity at the time of origination, and year. Figure 9 plots the estimates of separately for regressions where is either credit rating, unused loan commitment percentage, and indicator for equity market access. There is clear evidence of a strong pro-cyclical pattern. The sensitivity of refinancing likelihood to borrower fundamentals declines considerably during recessions. For example, refinancing likelihood becomes also insensitive to borrower fundamentals during the recession. 13 For example, Erel et al (2011) find that capital raising tends to be pro-cyclical for noninvestment-grade borrowers and counter-cyclical for investment-grade borrowers, and attribute this to the effect of macroeconomic conditions on the supply of capital. 27

28 Table 5 explicitly tests for the correlation between refinancing and borrower-quality sensitivity, and measures of credit cycle and business cycle strength (MACRO) by updating equation (5) to: 6 where MACRO is either measured by GDP growth or senior loan officer responses during year t. captures the correlation of interest, i.e. how the sensitivity of refinancing with respect to credit worthiness varies over the credit / business cycle. Columns (1) through (3) interact measures of credit worthiness with senior loan officer credit tightness index 14. The interaction coefficient is negative and statistically significant, implying that as credit conditions tighten refinancing becomes less sensitive to variables capturing credit worthiness including investment grade, availability of unused lines of credit, and access to equity markets. The magnitude of the interaction effect is very large as well. When credit conditions are very tight as in refinancing becomes almost insensitive to variables capturing credit worthiness. The three credit attributes used in columns (1) through (3) are positively correlated with each other. The pair-wise correlation between investment grade rating, and draw down percentage and publicly listed indicator variable is and respectively. The correlation between draw down percentage and publicly listed indicator variable is Thus while these variables are correlated as expected, the correlation is far from perfect. Column (4) includes all three of these variables simultaneously and shows that the interaction with each of the three remains significant and quantitatively strong. 14 The results are essentially identical if we use the senior loan officer survey response on credit spreads instead. 28

29 Column (5) repeats the regression but included bank-year fixed effects thus absorbing any variation driven by lead-bank specific annual shocks. The interaction coefficients remain largely unaffected. Bank-year fixed effects non-parametrically test if the cyclicality in refinancing sensitivity to borrower attributes was spuriously driven by the type of banks that firms borrow from. Column (6) replaces the MACRO variable with GDP growth that captures business cycle strength. All three interactions with GDP growth are significant. As always, the standard errors are clustered at the lead-bank level. The sensitivity of refinancing propensity is substantially more sensitive to credit and business cycles for credit-worthy firms for both early and at-maturity refinancing. Appendix figure 2 limits the focus to early refinancing and shows that more credit worthy borrowers both in terms of investment grade rating and having access to public equity see a sharper decline in early refinancing when credit conditions tighten. Columns (7) and (8) of table 5 repeat the tests of columns (5) and (6) after limiting the data to loans at-maturity and confirm that at-maturity loans are also more sensitive to credit market conditions for credit worthy borrowers. The fact that refinancing propensity of credit worthy borrowers is significantly more sensitive to credit market conditions suggests that the time-series reduction in refinancing propensity when credit conditions tighten is partly driven by demand-side response to higher cost of capital. First, active maturity management via early refinancing in normal times enables borrowers to delay refinancing when cost of capital rises, thus reducing the demand for refinancing. Second, higher cost of capital in tight credit market conditions dissuades more credit-worthy borrowers perhaps because of their deeper pockets from seeking to refinance their outstanding credit limits. 29

30 IV. Discussion of Alternative Interpretations Firms with better credit worthiness are more sensitive to credit market conditions in their refinancing probability. The cyclical variation in refinancing by credit worthy firms is driven by the proclivity to refinance early at a significantly higher rate in periods when credit conditions are loose. We have interpreted these core results as evidence in favor of active maturity management to minimize costs associated with liquidity risk. Could there be alternative explanations consistent with the collective evidence put together in this paper? There is little doubt that supply-side e.g. credit lending standards play an important role in generating the variation in liquidity risk over time. For example, evidence from loan officer surveys reported in figure 1 reflects supply-side changes in lending standards. However, in the absence of active maturity management, aggregate changes in supply-side conditions reflected in loan officer surveys are unlikely to generate the results we find. For example, as credit conditions tighten say in 1998 or the recession of banks will impose a harsher discipline on the marginal borrower first. These are likely to be less credit worthy borrowers that end up being the first to be denied credit, or given credit at harsher terms. Thus we would expect supply side changes to make the refinancing propensity of less credit worthy firms to be more sensitive to credit market conditions. We find completely the opposite. Our results suggest that there is a strong demand-side channel driven by maturity management effects that works in the opposite direction. Credit worthy firms deliberately refinance early at a higher rate when credit conditions loosen up, and choose to refinance less aggressively when credit conditions are tight. They can choose to do so due to longer effective 30

31 maturity, and possibly cheaper alternative sources of financing when cost of external financing goes up. An alternative demand-based explanation for our results could be that perhaps the natural demand for more credit worthy firms is more cyclical. Under this hypothesis, our results are not driven by some active maturity management by good quality firms. Instead they simply reflect the fact that refinancing needs are naturally more cyclical for credit worthy firms. While this view is hypothetically possible, we think it is unlikely since less creditworthy firms with uncertain cash flows are more susceptible to fluctuations in the business cycle. Thus under the standard demand-side view, we would expect credit worthy firms to be less sensitive to business cycle fluctuation. Nonetheless we show that when we explicitly control for business cycle variation the sensitivity of refinancing with respect to credit conditions remains as strong for credit worthy firms. V. Concluding Remarks The question of liquidity risk and maturity management in relation to the credit cycle holds a central place in financial economics. Yet most of the work in this area remains theoretical, primarily due to a paucity of data necessary to adequately answer the questions of interest. In this paper, we hope to have made an important contribution to this debate. The novel feature of our data is that we can directly observe the refinancing propensity of outstanding syndicate loans and the utilization rate of total loan commitments. An added advantage is the loan-level coverage of a wide cross-section of firms over twenty two years that include three important recessions and credit cycles. Our analysis reveals a strong relationship between refinancing likelihood, utilization rate of loan commitments and credit conditions. A striking feature of the refinancing behavior is the 31

32 tendency to favor early refinancing in normal times versus when credit conditions are tight. Such behavior enables firms to keep maturity long when credit is easy and hence have more ability to avoid having to refinance when credit terms are difficult. We find that credit worthy firms are best able to do so effectively. The net impact of more cyclical (with respect to the credit cycle) refinancing by better quality firms is that the type of firms that are most exposed to liquidity risk when credit conditions tighten are systematically weaker firms. There are a number of interesting and promising questions raised by our findings that we hope future scholars will take up. For example, this paper focused on how demand-side factors influence the refinancing choice and hence maturity structure. In terms of supply-side factors, our focus was to either control for possible bank-specific shocks through bank-year fixed effects, or to argue that the observed patterns are unlikely to be generated by supply-side forces. Of course bank specific supply-side shocks may have an independent effect on liquidity and maturity structure that is worth investigating in the future. 15 The secular decline in maturity of syndicate loans also warrants further investigation. One possibility is that increased reliance on short-term borrowing through the Repo market forced banks to issue less long-term loans. This is a question worth investigating in light of the debate on the consequences of shadow banking system on credit. Another question related to the shadow banking system is the role played by shadow financial institutions in the syndicate structure. Our data includes information on participant banks and can be used to investigate the type of investments favored by the shadow banking institutions. 15 Bord and Santos (2011) show that banks that were under more liquidity pressure following the collapse of the market for asset backed commercial paper in the fall of 2007 increased the cost they charge corporations for granting them access to liquidity. 32

33 On the supply side, the role of foreign financial institutions in possibly mitigating the adverse effects of domestic liquidity shocks can also be analyzed more carefully using the data analyzed in this paper. For example, did foreign lead banks step in to buffer liquidity in the syndicate market in the aftermath of the financial crisis? More generally, the microlevel detail of the data utilized in this paper offers exciting opportunities to more carefully understand the link between financial shocks, corporate financial policy and the real economy. 33

34 References Allen, F. and G. Gale, Understanding Financial Crises. Oxford University Press. Almeida, Heitor; V. Acharya and M. Campello, Aggregate Risk and the Choice between Cash and Lines of Credit, Working Paper, NYU. Barclay, M and Clifford Smith, The Maturity Structure of Corporate Debt, Journal of Finance 50(2), Berglof, E. and E. Von Thadden Short-Term Versus Long-Term Interests: Capital Structure with Multiple Investors. Quarterly Journal of Economics 109, Berger, A.N., M.A. Espinosa-Vega, W.S. Frame, and N.H. Miller Debt Maturity, Risk, and Asymmetric Information. Journal of Finance 60, Bodnar, Gordon, John R. Graham, Campbell R. Harvey and Richard C. Marston, Managing Risk Management, working paper. Bord, Vitaly, and João A.C. Santos, Bank Liquidity Standards and the Cost of Liquidity to Corporations Working paper, Federal Reserve Bank of New York. Brunnermeier, Markus K. and Martin Oehmke, The Maturity Rat Race, NBER Working Paper No DeMarzo, P., and M. Fishman Optimal Long-Term Financial Contracting, Review of Financial Studies 20 (6), Diamond, D., Debt Maturity Structure and Liquidity Risk, Quarterly Journal of Economics 106(3), Erel, Isil; Brandon Julio, Woojin Kim, Michael S. Weisbach, Macroeconomic Conditions and Capital Raising, Working Paper. Farhi, E., and J. Tirole Collective Moral Hazard, Maturity Mismatch and Systemic Bailouts, Mimeo, Harvard University and Toulouse School of Economics. Flannery, Mark J., Asymmetric Information and Risky Debt Maturity Choice, Journal of Finance 41(1), Froot. K.A., D.A. Scharfstein and J.C. Stein Risk Management: Coordinating Corporate Investment and Financial Policies, Journal of Finance 48(5), Gorton, G., and G, Pennacchi Financial intermediaries and liquidity creation, Journal of Finance 65(1), Griliches, Zvi and Jacques Mairesse 1995, Production Functions: The Search For Identification, NBER working paper #

35 Guedes, Jose and Tim Opler The Determinants of the Maturity of Corporate Debt Issues, Journal of Finance 51(5), Hart, O., and J.Moore A Theory of Debt Based on the Inalienability of Human Capital, Quarterly Journal of Economics 109, He, Zhiguo and Wei Xiong Rollover Risk and Credit Risk, Forthcoming Journal of Finance. Holmstrom, Bengt and Jean Tirole, Inside and Outside Liquidity, The MIT Pres. Ivanisha, V., and D. Scharfstein Bank Lending During the Financial Crises of 2008, Journal of Financial Economics 97, Johnson, Shane, Debt Maturity and the Effects of Growth Opportunities and Liquidity Risk on Leverage, Review of Financial Studies 16(1), Krishnamurthy, A., and A. Vissing-Jorgensen The Aggregate Demand for Treasury Debt, Working paper. Northwestern University. Khwaja, Asim and Atif Mian, Tracing the Impact of Bank Liquidity Shocks, American Economic Review 98, Roberts, Michael, The Role of Dynamic Renegotiation and Asymmetric Information in Financial Contracting, working paper. Roberts, Michael and Amir Sufi, Renegotiation of financial contracts: Evidence from private credit agreements, Journal of Financial Economics. Stohs, Mark and David C. Mauer, The Determinants of Corporate Debt Maturity Structure, Journal of Business 69(3), Santos, João A.C Bank corporate loan pricing following the subprime crisis, Review of Financial Studies 24(6), Santos, João A.C. and Andrew Winton Bank Loans, Bonds, and Information Monopolies across the Business Cycle, Journal of Finance 63(3), Sufi, A Information Asymmetry and Financing Arrangements: Evidence from Syndicated Loans, Journal of Finance 62,

36 Figure 1 Loan Officers Survey Results The top-left panel plots senior loan officer loan tightening survey for large and medium C&I loans. The top-right panel plots senior loan officer increasing spread survey for large and medium loans. The bottomleft panel plots bank charge-offs over assets (weighted and un-weighted). The bottom-right panel plots corporate bond spreads over time. Vertical dashed lines represent NBER-dated recessions. loan officer survey tightening for large & medium C&I? loan officer survey increasing spreads for large & medium? jan199001jan199501jan200001jan200501jan jan199001jan199501jan200001jan200501jan2010 bank chargeoffs over assets AAA-BAA 1 year spread q1 1995q1 2000q1 2005q1 2010q1 01jan199001jan199501jan200001jan200501jan2010

37 Figure 2 Frequency Distribution Of Credit Variables This figure plots the frequency distribution for some key variables in our data. A unit of observation is a syndicate loan observed as of December 31 st of each year. The data cover period from 1988 to Change in maturity refers to the change in maturity date of a loan (if any), Maturity left measures the days left till maturity of a given loan in our sample, Maturity at origination refers to the maturity of a loan at the time it was originated, Draw down percentage refers to the percentage of total loan commitment currently drawn down by the borrowing firm.

38 Figure 3 Propensity to Refinance This figure plots the average propensity for syndicate loans to be refinanced in a given year. A loan is assumed to have not been refinanced if it does not appear in the sample in the subsequent year. Vertical dashed lines represent NBER-dated recessions.

39 Figure 4 Intensive Liquidity: Draw Down Percentage The top panel plots the average drawn down percentage (loan outstanding divided by loan commitment) for syndicate loans over time. The lower plans plot the percentage of loans that are fully drawn down. Vertical dashed lines represent NBER-dated recessions.

40 Figure 5 Effective Maturity Over The Business Cycle The top panel plots the average maturity left of all outstanding syndicate loans at the end of a year. The bottom panel plots the fraction of syndicate loans that are maturing within a year. Vertical dashed lines represent NBER-dated recessions. Effective Maturity (in years) Fraction of loans At-Maturity

41 Figure 6 Propensity to Extend Maturity And Maturity Left The top panel (left axis) plots the propensity to refinance a loan over the 1988 to 2010 period against the number of years left until the expiration of the existing loan. 0 means the loan has less than one year left till maturity, 1 means the loan has between 1 and 2 years left till maturity, and so on. 10+ means the loan has ten or more years left till maturity. The right axis plots the share of total refinancings that belong to refinancings with x years left till maturity. The blue (dotted) line in bottom panel plots the estimated coefficients from the OLS regression:, where Y is an indicator variable for whether the syndicate loan i of borrower j from lead bank b is refinanced during year t (0/1). is the number of years till maturity fixed effect that turn 1 if the number of years left till maturity is between n and n+1, where n varies from 0 to 10. All loans above 10 years left in maturity are top-coded at 10. is the number of years for maturity as of origination fixed effects (also going from 0 to 10). is the year fixed effects. The red (dashed) line plots the same coefficients, but includes borrower-year ( ) fixed effects, thus comparing loans of differing maturities within the same firm-year.

42 Figure 7 Probability of Early Vs. At-Maturity Refinancing The top panel plots refinancing likelihood for loans maturing within the next year (blue-dashed line and left axis), and for loans maturing in more than one year (red solid line and right axis). In the bottom panel, the blue line plots the estimated coefficient ( of the following logit model separately for each year: Pr 1, where Y is an indicator variable on whether the syndicate loan i of borrower j from lead bank b is refinanced during year t (0/1). The red line limits the sample to loans of differing maturities issued to the same firm in a given year. Vertical dashed lines represent NBER-dated recessions.

43 Figure 8 Maturity Refinancing By Loan Type The figure plots the maturity-adjusted propensity to refinance over time for various category of firms. The maturity-adjustment is done by taking the residual after regressing the incidence of refinancing (0/1) on fixed effects for number of years of maturity at origination and fixed effects for number of years left till maturity at the time of observation. The top panel plots the maturity-adjusted refinance probability for loans that are less than 5% drawn down, and loans with more than 95% drawn down. The middle panel plots the same propensity for investment grade and non-investment grade loans respectively, and the lower panel for publicly-traded and non-publicly traded firms. Vertical dashed lines represent NBER-dated recessions.

44 Figure 9 Dependence Of Refinancing On Borrower Attributes Over Time This figure plots the from the regression:,, where Y is an indicator variable for whether the syndicate loan i of borrower j from lead bank b is refinanced during year t (0/1) and X captures borrower attributes as of (t-1). We plot for three X s: indicator variable for investment grade rating, unused loan commitment (i.e. one minus draw down percentage), and indicator variable for weather borrow is publically listed. A separate regression is run every year, and each regression includes number of years left till maturity fixed effects ( ), number of years for maturity as of origination fixed effects (, and year fixed effects ( ). Vertical dashed lines represent NBER-dated recessions.

45 Appendix Figure 1 Time Series Of Corporate Liquidity The top panel plots the average maturity extension conditional on refinancing. The bottom panel plots the share of total loans in a given year that are new originations. Vertical dashed lines represent NBER-dated recessions.

46 Appendix Figure 2 Early Refinancing By Borrower Type This figure presents results of regressing whether a syndicate loan gets refinanced in year t (0/1) on variables capturing borrower quality and fundamentals as of (t-1). A separate regression is run every year, and each regression includes lead bank fixed effects, number of years left till maturity fixed effects, and number of years for maturity as of origination fixed effects. Vertical dashed lines represent NBER-dated recessions.

47 Table 1 Data Tabulation This table presents the tabulation of data over time and across some categories of interest. The original SNC data is at the level of a syndicate loan, and tracks each loan over time. Information on each loan is provided as of December 31 st of each year. Column (1) reports number of loans each year in our data, column (2) reports the number of borrowers (firms) each year, column (3) reports the sub-sample of firms in (2) that borrow from more than one leadbank in a given year, column (4) reports the sub-sample of firms in (2) that have multiple loans such that one loan is maturing in the current year, and another is maturing later in time. Column (5) reports the number of lead banks by year, and column (6) breaks down syndicate loan panel by industry. (1) (2) (3) (4) (5) (6) Year # of loans # of firms # of firms with multiple banks # of firms with multiple maturities # of lead banks # of loans Industry Agriculture- Mining 7, ,309 3, ,853 3, Utilities 11, ,332 4, Construction 11, ,221 3, Manufacturing 46, ,522 3, Trade 19, ,643 3, Transport 5, ,397 4, Information 12, ,026 4, Real Estate 11, ,576 4, Services 21, ,745 5, Unknown 8, ,539 4, ,376 4, ,463 4, ,880 4, ,194 4, ,428 3, ,952 3, ,141 4, ,711 4, ,603 4, ,809 5, ,280 4, ,041 5, Total 156, ,322 4,217 5,749 4,606 Total 156,041 (50,469 unique loans) (22,165 unique firms) (661 unique banks)

48 Table 2 Summary Statistics This table presents summary statistics for the Shared National Credit (SNC) program data on syndicate loans. The data track each loan over time. Information on loans is provided as of December 31 st of each year from 1988 to The top panel presents summary statistics for loan level data. %age of shadow bank participation is the percentage of loan that is funded by non-commercial banks, or financial firms outside of the Fed banking supervision. Refinanced? equals 1 if a loan is refinanced (or rolled over) in a given year. The middle panel reports summary statistics on borrowing firms, but the sample is limited to borrowers that are publically listed on a stock exchange. The bottom panel reports summary statistics on three macroeconomic variables. N Mean SD 10 th 50 th 90 th Loan level data Total Commitment 156, , ,005 15,000 73, ,106 Total Outstanding 156,039 81, , , ,741 Draw Down Percentage 156, Investment Grade? 156, Non Accrual? 71, Lead Bank Share 156, Foreign Lead Bank? 156, %age Shadow Bank Participation 155, Refinanced? 144, Change in Draw Down Percentage 102, Firm level data Total Assets 24,615 3,544 18, ,724 Log Sales 24, Return on Assets 24, Total Sales 24,615 2,409 9, ,287 Sales on Assets 23, Macro variables data GDP Growth SLOOS Tightening SLOOS Spread

49 Table 3 Liquidity And Borrower Fundamentals This table presents and from the regression:,,, where Y is an indicator variable for whether the syndicate loan i of borrower j from lead bank b is refinanced during year t (0/1). X and Z capture loan and borrower characteristics respectively as of (t-1). Beginning in column (2), separate fixed effects for year, number of years left in maturity, number of years for maturity at origination, and industry are incorporated. Bank-year fixed effects are included from column (4) onwards. A unit of observation is a syndicate loan, and data cover a period from 1988 to The estimation procedure is OLS (linear probability), and standard errors are clustered at the lead bank-year level (average of 221 lead banks over 22 years). RHS variables as of (t-1) Dependent Variable: Loan Refinanced At Time t? Less than one year left in maturity (1) (2) (3) (4) (5) (6) (7) Investment Grade? 6.68** 6.58** 9.15** 6.61** 9.87** 13.53** 18.30** (0.446) (0.434) (1.094) (0.456) (1.236) (1.385) (4.279) Draw Down (%) -1.35* (0.629) (0.502) (0.960) (0.500) (0.996) (1.732) (4.172) Draw Down > 95%? -2.58** -3.52** -5.67** -2.87** -5.75** -8.62** (0.503) (0.403) (0.965) (0.419) (1.031) (1.542) (4.630) Log change in sales 1.32* 1.28* 1.70 (0.576) (0.610) (2.574) Return On Assets (0.013) (0.012) (0.365) Sales over Assets (0.043) (0.046) (0.159) Credit Line? 2.60** 2.82** ** ** 8.27** (0.496) (0.350) (0.770) (0.374) (0.829) (1.212) (2.892) Publicly Listed? 4.11** 5.42** 5.08** 4.83** (0.420) (0.353) (0.358) (1.181) Maturity Left (in yrs) -4.58** (0.107) Maturity at Origination (yrs) -0.88** (0.084) Bank-Year Fixed Effects yes yes yes yes Industry Fixed Effects yes yes yes yes yes yes # of Years of Maturity Left yes yes yes yes yes yes Fixed Effects Year Fixed Effects yes yes yes yes yes yes # of Years of Maturity At yes yes yes yes yes yes Origination Fixed Effects N 147, ,454 36, ,454 36,230 21,032 4,480 R **,*,+ Coefficient statistically different than zero at the 1%, 5%, and 10% confidence level, respectively

50 Table 4 Estimate Of Early Refinancing This table presents and from the logit model: Pr 1, where Y is an indicator variable for whether the syndicate loan i of borrower j from lead bank b is refinanced during year t (0/1). LM is an indicator variable for whether the loan has more than one year left till maturity. is a measure of either credit conditions or business cycle strength during year t. Credit conditions are measured by senior loan offers surveys and business cycle strength is measured by average GDP growth. Fixed effects (Γ ) are specified in each column. The coefficient ( ) on LM thus captures the log of the odds ratio for early versus at- f maturity refinancing (i.e. ln ). Column (7) limits the sample to loans of differing f maturities issued to the same firm in a given year. A unit of observation is a syndicate loan, and data cover a period from 1988 to Standard errors are clustered at the lead bank-year level (average of 221 lead banks over 22 years). RHS variables as of (t-1) Loan Refinanced At Time t? (1) (2) (3) (4) (5) (6) (7) Long Maturity (LM)? -1.36** -1.31** -1.27** -1.36** -1.59** -1.33** -1.26** (0.028) (0.033) (0.030) (0.029) (0.045) (0.087) (0.056) LM * 1990 Recession 0.06 (0.097) LM * 2001 Recession -0.28** (0.076) LM * Recession -0.51** (0.115) LM * SLOOS Tightening -0.90** -0.80** -1.17** (0.099) (0.194) (0.255) LM * SLOOS Spread -0.55** (0.062) LM * GDP 0.09** 0.02 (0.016) (0.029) Industry Fixed Effects yes yes yes yes yes yes yes Year Fixed Effects yes yes yes yes yes yes yes # of Years Of Maturity At Origination yes yes yes yes yes yes yes Fixed Effects N 147, , , , , ,465 15,138 **,*,+ Coefficient statistically different than zero at the 1%, 5%, and 10% confidence level, respectively

51 Table 5 Liquidity And Borrower Fundamentals Over The Business Cycle This table presents and from the regression:,, Γ, where Y is an indicator for whether the syndicate loan i of borrower j from lead bank b is refinanced during year t (0/1). Borrower fundamentals attributes, are measured as of (t-1), and include indicator variable for investment grade rating, percentage loan commitment drawn down, and indicator variable for borrower being listed on the stock exchange. is a measure of either credit conditions or business cycle strength during year t. Credit conditions are measured by senior loan offers surveys and business cycle strength is measured by average GDP growth. Fixed effects (Γ ) are specified in each column. A unit of observation is a syndicate loan, and data cover a period from 1988 to The estimation procedure is OLS (linear probability), and standard errors are clustered at the lead bank-year level (average of 221 lead banks over 22 years). Less than one year left in maturity Loan Refinanced At Time t? (1) (2) (3) (4) (5) (6) (7) (8) RHS variables as of (t-1) Investment Grade? 10.62** 8.49** 8.68** 3.05** 16.77** 10.31** (0.48) (0.47) (0.49) (0.62) (1.40) (2.67) Investment Grade * SLOOS Tightness ** ** ** (1.71) (1.62) (1.65) (6.29) Draw Down (%) -8.08** -5.27** -4.93** -2.99** ** -4.55* (0.37) (0.37) (0.36) (0.48) (1.50) (2.25) Draw Down * SLOOS Tightness 10.14** 6.40** 6.46** 15.8** (1.55) (1.45) (1.44) (5.86) Publicly Listed? 7.65** 6.27** 5.77** 3.75** 3.29** 5.90** (0.40) (0.39) (0.40) (0.50) (1.27) (2.21) Publicly Listed * SLOOS Tightness ** -8.34** -6.95** 9.53*** (1.56) (1.48) (1.48) (5.02) Investment Grade * GDP Growth 1.58** 1.84* (0.23) (0.86) Draw Down * GDP Growth -0.49** -1.96** (0.16) (0.73) Publicly Listed * GDP Growth 0.52** (0.17) (0.65) Bank-Year Fixed Effects yes yes yes yes Industry Fixed Effects yes yes yes yes yes yes yes yes Year Fixed Effects yes yes yes yes yes yes yes yes # of Years Of Maturity At Origination Fixed Effects yes yes yes yes yes yes yes yes # of Years Of Maturity Left Fixed Effects yes yes yes yes yes yes yes yes N 143, , , , , ,168 20,152 21,031 R-sq **,*,+ Coefficient statistically different than zero at the 1%, 5%, and 10% confidence level, respectively

52